Expansion From Lowland Pleistocene Refugia Followed by Cryptic Introgression in an African Elevational Generalist Bird

ABSTRACT

The immense biodiversity observed in tropical forests has inspired the study of evolution since its inception, and numerous hypotheses have been generated to explain this diversity. One major driver of tropical forest biodiversity has been the repeated climatic fluctuations due to the Pleistocene glacial cycles. The Afrotropics are one region that has seen a particularly rich debate about the origin of its avifauna. Lowland and highland Afrotropical forests responded in different ways to Pleistocene climatic cycling, likely shaping the population genetic structure of species tracking these habitats. Here we study the role of shifting forests throughout the Pleistocene in shaping population genetic structure in a forest generalist, the Yellow‐whiskered Greenbul ( Eurillas latirostris ). We sequenced genomes from individuals across the continent with forest thought to span three Pleistocene refugia. At a coarse scale, we found evidence for three lineages that expanded from these refugia, with a western lineage west of the Dahomey gap, a central lineage across the Congo Basin, and an eastern lineage in the Albertine rift and adjacent rainforest. However, finer scale analysis revealed a more complex story. First, we found that the most stable eastern lineage demonstrated population genetic structure between its montane populations. This included one population that was likely admixed between the sampled Albertine Rift locations and an unsampled region. Second, we found surprisingly strong divergence between the parapatric eastern and central lineages. However, despite this divergence, we found signals of mitochondrial introgression from the eastern lineage far into the eastern sampling localities of the central lineage. This suggests a largely untapped promise for evolutionary genomic research across the Congo Basin, where populations expanding from alternate refugia likely meet and have their species boundaries tested.

1. Introduction

The field of phylogeography has given evolutionary biologists a strong sense of how genetic diversity is distributed across the landscape, revealing how ecological processes, barriers to dispersal, and expansion from Pleistocene refugia shape observed patterns across species (Pârâu and Wink 2021; Edwards et al. 2022; Fu and Wen 2023). The deviations from continuous isolation‐by‐distance patterns (i.e., phylogeographic breaks; Swenson and Howard 2005; Edwards et al. 2022) often reflect these different processes, either manifesting across presently continuous (Irwin 2002) or discontinuous (Thom and Aleixo 2015; Musher 2025) suitable habitat. Phenotypically cryptic generalist taxa have often been overlooked in phylogeographic scale studies—especially in vertebrates—due to their assumed panmixia (Marks 2010; Williamson et al. 2024). However, some generalists are still restricted to broad habitat types such as forest generalists; these taxa may occupy a wide variety of forested habitats or forests across a broad elevational range. Generalists such as this can experience parallel range expansions and contractions with the specialists they co‐occur with. For forest generalists, geographic range shifts could occur due to consequences of Pleistocene glacial cycles such as range retractions from glaciers or changes in aridity (Morales‐Rozo et al. 2017) and range expansions from shifting sea levels (Brown et al. 2013). However, they do not track habitat as strictly as specialists, allowing for fewer assumptions about past ranges.

One classic driver of phylogeographic structure is expansion from Pleistocene refugia, either caused by glaciers wiping away potential habitat or major climatic shifts (Provan and Bennett 2008; Androski et al. 2024). Due to the recency of these changes relative to the effective population size of many species, Pleistocene dynamics are often well represented in present genetic structure (Knowles and Richards 2005; Smith et al. 2022). Indeed, modern genetic structure may not be an equilibrium state, depending on the precise characteristics of the species (Barton and Hewitt 1985; Luttikhuizen et al. 2018). This possible lack of equilibrium allows us to test how factors like demographic expansions shape secondary contact in lineages that diverged during vicariance events caused by Pleistocene dynamics. Understanding how these expansions occur is essential to conserving species' genetic diversity and understanding how future global change induced range shifts impact these species' evolution in the future.

One critical process facilitated by range expansion is secondary contact of incipient species, strongly divergent populations that were once isolated (Excoffier et al. 2009; McGaughran et al. 2022; Wait and Peñalba 2025). This can manifest in explicit parapatry or in increased proximity, facilitating gene flow. Although taxa impacted by Pleistocene fluctuations are likely not experiencing their first instance of contact, it is often their first opportunity to exchange genes in tens to hundreds of thousands of years (Papadopoulou and Knowles 2015; He et al. 2019). This is where potential reproductive boundaries and ecological divergence are tested, and can result in a wide array of outcomes, from completion of speciation to genetic swamping (Abbott et al. 2013; Kulmuni et al. 2020; Shogren et al. 2024). We can attempt to understand the role of different barriers in shaping the future of lineages based on the observed patterns of secondary contact where those lineages meet, which can leave traces across the landscape depending on the exact history (Andersen et al. 2021). One important factor in this is the role of differences in population expansion rate, which can result in the net expansion of a lineage over time, or even genetic swamping in cases of asymmetric population size (Excoffier et al. 2009; Gyllenhaal et al. 2025). Phylogeographic data can show interesting patterns such as those predicted by the “wavefront” model of introgression, where population expansions can be reflected in modern distributions of regions resistant to introgression (e.g., mitochondria; Petit and Excoffier 2009; Rheindt and Edwards 2011).

The Afrotropics are one region where cyclic expansions and contractions of forested habitat have been hypothesized to shape biodiversity and the distribution of genetic variation on the landscape. Three models have dominated discussions looking to understand the governing mechanisms of species diversification in the Afrotropics, aiming to explain the radiation of lowland and highland taxa in the region. The original model (Chapin 1932; Moreau 1954; Fjeldså 1994) evokes natural selection acting across ecotones (habitat transitions) along the periphery of the lowland rainforest as the driving force for species diversification. This model was first proposed as a mechanism to explain the dramatic species turnover across the transition from savanna to rainforest habitats (Chapin 1923, 1932) and lowland to montane rainforest habitats (Moreau 1954). This model has been proposed at the intraspecific level in Central African populations of the Little Greenbul ( Eurillas virens ), whereby populations show morphometric and genetic divergence across this savanna–rainforest ecotone (Smith et al. 1997, 2005; Zhen et al. 2017).

The additional models have since been proposed that centered on the role of changes in habitat extent over time. The first one focuses on lowland areas forest taxa and the role of cyclic bouts of allopatry as climate fluctuations produced variable habitat over time (Haffer 1969 for South America; Diamond and Hamilton 1980; Crowe and Crowe 1982; Mayr and O'Hara 1986, for Africa). This model posits that periods of increased global cooling led to reduced rainfall in tropical areas, causing contraction of widespread lowland forests into refugial islands of suitable habitat and allowing for allopatric divergence, followed by secondary contact during warm periods (Mayr and O'Hara 1986; Maley 1991). The second refugial model focused on range‐restricted species in highlands that shared affinities across wide areas of unhospitable rainforest, while lowland species were often phenotypically undifferentiated species across extensive lowland ranges (Moreau 1954). This observation led to the hypothesis that stable montane habitat acted as a source for lineages to accumulate in the lowlands despite a lack of recent diversification (Fjeldså and Lovett 1997; Fjeldså et al. 2007). Many widespread lowland forest species are indeed morphologically homogeneous (Diamond and Hamilton 1980; Mayr and O'Hara 1986), but phylogeographic studies of many widespread lowland rainforest species globally have demonstrated that species which appear outwardly uniform often comprise a remarkable degree of geographically structured genetic diversity (e.g., Zou et al. 2007 in Southeast Asia; Marks et al. 2002; Thom et al. 2015 in South America; Marks 2010; Huntley et al. 2019 in Africa; Miura and Edwards 2001 in Australia). This suggests that cryptic taxa may be masking deeper evolutionary histories in the lowlands.

The bulbuls and greenbuls in the family Pycnonotidae are one group of birds that has seen notable attention in the pursuit of understanding the processes of diversification in African rainforests. Within this family, one genus, Eurillas, has played an especially important role. Two of the three main models of Afrotropical diversification were tested empirically through intraspecific comparisons across different Eurillas species, particularly the role of natural selection across the forest savanna ecotone (Smith et al. 1997, 2005; Zhen et al. 2017) and the role of genetic drift in isolated montane regions (Roy 1997). Here we assess the montane and lowland refugial hypotheses of the Afrotropics via genomic phylogeographic analysis of a generalist member of the genus Eurillas: the Yellow‐Whiskered Greenbul ( Eurillas latirostris ).

Eurillas latirostris is the most widespread of the rainforest greenbuls, occurring from Sierra Leone in the west, east to southeastern Tanzania, and from South Sudan to southern Democratic Republic of Congo (DRC). Frequently heard, but rarely seen, it is abundant throughout its range and is a forest elevational generalist, occurring in primary and secondary forests from sea level at the Gulf of Guinea to more than 2700 m in the Rwenzori Mountains of the Albertine Rift. In forests, it inhabits every stratum from the floor to the canopy. It is an omnivore, feeding on vertebrates, invertebrates, seeds, and a variety of fruits (Brosset 1982). There are five described subspecies, with three widely recognized (AviList Core Team 2025): one in the west African forests from Senegal to Nigeria (E. l. congener), a second from southwestern Nigeria through western Kenya and south to Angola (E. l. latirostris), and a third inhabiting the Ufipa Plateau in Tanzania (E. l. australis). We focus on two of these subspecies (latirostris and congener) to assess the role of stable mountains and dynamic lowlands in shaping population structure in African rainforests. Broad geographic and elevational sampling in this study encompasses three hypothesized lowland refugia—from west to east, Upper Guinea, Western Lower Guinea, and Eastern Lower Guinea—and an elevational gradient in East Africa (Albertine Rift). This provides a rare opportunity to assess the relative roles played by lowland versus montane regions and isolation by distance or elevational barriers in shaping avian diversity in African rainforests. We test two fundamental hypotheses: that lowland rainforest refugia correlate with genetic structure in this generalist, and that mountain populations will show weaker signatures of post‐glacial population expansion than lowland ones. Additionally, we will test if the generalist nature of this taxon results in panmixia among mountain populations in contrast with the strong structure shown by montane specialists.

2. Methods

2.1. Sampling, Laboratory Work, and Sequencing

We obtained tissue samples for 55 E. latirostris individuals from across its geographic distribution from natural history museums (Figure 1A and Table S1). In addition, we received tissues for two E. virens individuals to use as outgroups. We extracted genomic DNA from tissue samples using the QIAGEN (Hilden, Germany) DNeasy blood and tissue kit following manufacturer guidelines. We sent these extracts to the Texas Tech University Center for Biotechnology and Genomics, where whole genome resequencing libraries were created using the standard Illumina DNA Prep kit by the center following their standard protocols. The libraries were then multiplexed and sequenced on two lanes of an Illumina NovaSeq6000 S4 flow cell. We aimed for equal representation of these samples on the sequencing lanes.

FIGURE 1.

Sampling and population structure of Yellow‐whiskered Greenbul ( Eurillas latirostris ). (A) Map of sampling localities, coloured by nuclear genetic cluster. Inset is an illustration of study species (art by AAR). (B) Admixture plot of all samples, roughly from west to east, labelled by sampling locations (DRC, Democratic Republic of the Congo; CAR, Central African Republic). (C) Genomic principal component analysis of all samples, coloured by sampling region. Note that the slightly diverged eastern (yellow) points correspond to northeast Uganda samples.

2.2. Reference Genome

Due to the fragmented nature of the existing Eurillas latirostris reference genome (GCA_041108415.1; Murhabale Cisirika et al. 2025), we performed reference‐based correction and scaffolding using RagTag v2.1.0 with the chromosome‐level Sylvia atricapilla reference genome (GCF_009819655.1) as a guide. We chose this reference genome as a guide because there were no chromosome scale reference genomes in the Pycnonotidae, and the Sylviidae family is in the clade sister to the Pycnonotidae (Oliveros et al. 2019). First, we used RagTag's “correct” module with default settings to break the scaffolded input reference at gaps to account for assembly errors. Then, we used RagTag's “scaffold” module with default settings to align the sequences of the contigs to the chromosome‐level reference. We discarded unplaced sequences from the reference. This produced a reference with sequence from Eurillas latirostris and chromosomal placement of the reference Sylvia atricapilla .

2.3. Quality Control and Alignment of Sequence Data

For five individuals, library preparations or sequencing failed, and we retained 50 ingroup individuals. We trimmed and filtered the raw sequencing data using BBDuk (Bushnell 2014). We then used BWA‐mem v2.2.1 (Li and Durbin 2009) to align the filtered reads to our corrected and scaffolded reference genome. After alignment, we processed the output SAM file with samtools v1.6 (Li et al. 2009) and Picard Tools (McKenna et al. 2010) to convert to BAM format, clean and sort the files, and remove duplicates. We also used samtools to quantify the mean genomic sequencing coverage per individual. One additional individual had very low coverage and number of genotyped sites and was removed from nuclear (but not mitochondrial) analyses, resulting in our final set of 49 ingroup individuals (Table S1).

2.4. Genotyping and Genotype Filtering

Using the cleaned and filtered BAM files, we genotyped each individual with the mpileup command of bcftools v1.17 (Li 2011), retained sites with a minimum of six sequencing reads, and merged all individuals' variant calls into variant call format (VCF) using the bcftools merge command. We further filtered these combined VCF files using vcftools v0.1.16 (Danecek et al. 2011) with various settings for different downstream analyses (Table S2).

2.5. Genetic Structure

We estimated the genome‐wide pattern of phylogeographic structure using three methods. First, we used PLINK v1.90b6.26 (Chang et al. 2015) to run a principal components analysis (PCA) on the SNP data. Second, based on visualization of the PCA output (Figure 1C), we ran ADMIXTURE v1.3.0 (Alexander and Lange 2011; Alexander et al. 2009) with an assumed number of genetic clusters from two to five. To quantify the level of divergence between populations identified with these analyses, we also calculated the weighted mean of genome‐wide Reich's F_ST (Reich et al. 2009) based on windowed values (see Population genetic summary statistics).

2.6. Phylogenetics

To visualize relationships between individuals and detect patterns of reticulation, we computed an unrooted phylogenetic network using the Neighbour‐Net method in SplitsTree v4.18.2 (Hudson and Bryant 2006). As input, we used a matrix of pairwise Nei's D (Nei 1972) distances calculated in the R package StAMPP (Pembleton et al. 2013).

To further infer the phylogenetic relationships among sampled individuals, we estimated gene trees for 100‐kbp non‐overlapping windows and retained only windows containing a minimum of 10,000 genotyped sites. For each retained window, gene trees were inferred using RAxML v8.2.12 (Stamatakis 2014) under the GTRGAMMA model of sequence evolution. This generated 9170 gene trees, which were later summarized into the best‐supported tree using the sumtrees.py script, a part of the DendroPy 4.6.1 Python package (Sukumaran and Holder 2010). Finally, we used RAxML gene trees as input for ASTRAL III v 5.7.1 (Zhang et al. 2018) using default settings to generate a species tree.

2.7. Mitochondrial DNA Analysis

We extracted the NADH dehydrogenase 2 (ND2) gene from all individuals by implementing the MitoFinder v1.4.2 pipeline (Allio et al. 2020) and using Megahit v. 1.2.9 (Li et al. 2016) as the assembler. Mitochondrial contig identifications were then performed with BLAST using the ND2 gene sequence from GenBank for E. virens (MG762099.1) as a reference. We implemented a phylogenetic approach to further investigate evolutionary relationships across E. latirostris populations over their geographic range. First, we aligned the ND2 sequences using MAFFT v. 7.511 under the method L‐INS‐i (Katoh and Standley 2013). Maximum‐likelihood phylogenetic trees were constructed with IQ‐TREE v1.6.12 (Nguyen et al. 2015) and ultrafast bootstrap (1000 replicates). The best substitution model selected by ModelFinder as part of the IQ‐Tree package (v1.6.12) (Kalyaanamoorthy et al. 2017) was TN + F + I (Hoang et al. 2018). We also constructed a median‐joining network of ND2 haplotypes with Hapsolutely (Vences et al. 2024).

2.8. Population Genetic Summary Statistics

To evaluate patterns of nucleotide diversity (π), heterozygosity (Ho), divergence (D_XY), genetic differentiation (F_ST), and allele frequency (Tajima's D) between genetic clusters of E. latirostris and determine whether these patterns vary across genomic regions, we used custom R scripts to calculate and plot statistics from non‐overlapping 100 kbp genomic windows. These and other custom scripts used here can be found on GitHub (https://github.com/ethangyllenhaal/YellowWhiskeredGreenbul/) and Dryad (https://doi.org/10.5061/dryad.b8gtht7sx). For F_ST and D_XY, these included comparisons between the three main genetic clusters we recovered as well as between more fine‐scale divergence revealed by additional analyses. Genome‐wide weighted mean values of these statistics were calculated based on the number of variable sites for F_ST and the total number of sites for other metrics. Additionally, due to the relatively high prevalence of females in our dataset, we performed analyses with all individuals without the Z chromosome, and analyses for only males with the Z chromosome (although, notably, we lacked females from the western populations, Table S1).

2.9. Demographic Histories

To understand the history of divergence in the context of gene flow, we used GADMA v2.0.0 (Noskova et al. 2023), a site frequency spectrum (SFS) based demographic modelling framework. We generated 3‐dimensional SFS two ways, the first using all variant autosomal sites with a maximum of 10% missing data. To better account for linkage disequilibrium, we also made an SFS by first thinning all sites (variant and invariant) by 1000 bp, then using the variant sites as input for the SFS. Both excluded two samples from northeast Uganda, see Results. A downsampled SFS was computed using easySFS v0.0.1, with projection number chosen to maximum the number of segregating sites (https://github.com/isaacovercast/easySFS; using code from dadi, Gutenkunst et al. 2009). We ran GADMA using a structured model that automatically finds the best fit demography, with moments v1.2.2 (Jouganous et al. 2017) as the underlying engine. Total sequence length was estimated using the sum of variant and invariant sites with the same filtering parameters, mutation rate was 4.6 × 10⁻⁹ (Smeds et al. 2016), and time per generation was 3.27 years (Bird et al. 2020). One demographic event was allowed per interval, the maximum effective population size was set to 10 times the ancestral size (based on preliminary runs), and the best run (chosen by likelihood) was chosen from 20 replicates run in parallel.

2.10. Fine‐Scale Population Structure

To better understand the role that landscape features play in shaping population structure, we performed additional analyses both at a species level and within the populations sampled in the Albertine Rift. At the species level, we visualized the geographic context of population structure with the program Estimated Effective Migration Surfaces (EEMS; Petkova et al. 2016). After generating input with PLINK v2.00a5.12 (Chang et al. 2015) and EEMS's bed2diffs program, we ran 20 replicate EEMS runs with different seeds, 5 million MCMC iterations (discarding the first 1 million as burn‐in), and a deme size of 1000. MCMC chains were visualized to confirm convergence and finally proceeded to plot effective migration and diversity surfaces using reemsplot2.

Additionally, we investigated population structure within major genetic clusters to determine if the mountain (eastern) populations displayed stronger structure than lowland (central and western) ones. For this, we first ran a principal component analysis with adegenet v2.1.10 (Jombart 2008) using an input generated with vcfR v1.15 (Knaus and Grünwald 2017). Next, we evaluated the effects of geographic distance and local adaptation on the genetic structure of the lowland and mountain eastern populations by implementing a model‐based approach for IBD (isolation by distance) and IBE (isolation by elevation). For both methods, we used as the source of genetic information a matrix of pairwise Nei's D distances computed from a subset of 15 individuals, which represent an altitudinal gradient between 1161 and 2629 m. An equivalent matrix of pairwise physical geographic distances in kilometres was calculated using the Haversine formula, as implemented in the R package fossil v0.4.0 (Vavrek 2011). Finally, as a proxy of differential local environmental conditions, we estimated a matrix of absolute pairwise altitude distances using base functions from R. The three resulting matrices were then integrated into a model framework through the multiple regression on distance matrices function from the R package ecodist v2.1.3 (Goslee and Urban 2007). The genetic distance matrix was used as the response variable, whereas the geographic and altitude matrixes were used as explanatory variables; significance was evaluated with 10,000 permutations. Additionally, we performed and plotted independent linear regressions between the genetic distances and the geographic and altitude distances to better visualize the relationships between variables.

2.11. Tests of Gene Flow

To measure the extent of gene flow among E. latirostris genetic clusters as well as ancestral gene flow that occurred prior to divergence between these clusters, we used a combination of custom R scripts and the DSuite v0.5r44 software package (Malinsky et al. 2021) to calculate whole‐genome D and f_d statistics. To begin, we calculated whole‐genome D statistics between two eastern populations (Rwenzori and the true East), two central populations (Central and Central core), and a western population of E. latirostris using DSuite. We also repeated this measure while considering northeastern Ugandan samples as a separate population. D statistics (also known as ABBA‐BABA tests) determine the frequency of ancestral (A) versus derived (B) alleles in asymmetric four‐taxon phylogenies including an outgroup (here, we used E. virens ). However, D statistics may be misleading when gene flow involved one or more internal branches of a phylogeny, and we accounted for this by using DSuite to calculate a genome‐wide f_branch metric that uses f₄ admixture ratios (Malinsky et al. 2021) on all branches of a given phylogeny and quantifies proportions of the genome that are shared between populations due to introgression.

Additionally, to understand the impact major admixture events might have on phylogenetic inference, we used TreeMix v1.13 (Pickrell and Pritchard 2012) to infer a population tree with admixture events. We added migration events until the last major change in likelihood occurred (following Fitak 2021). We added an additional filter of a minimum minor allele count of 5 to ensure variation within the ingroup was well‐represented (4 included fixed variants between E. virens and latirostris).

3. Results

3.1. Dataset Characteristics

We aimed to sequence each individual with 10–20× genomic coverage. Unfortunately, our data had a moderate amount of adapter contamination and 30%–40% PCR duplicates per individual. Because of these issues, our filtered dataset resulted in 6.8–15.3× genomic coverage per sample and 461–1009 million genotyped sites (after coverage filtering) per individual (Table S1). Characteristics of datasets are in Table S2, but all contained a high volume of information sufficient for downstream analyses.

3.2. Genetic Structure

Both PCA and ADMIXTURE results (Figure 1) indicated strong sorting of individuals into three phylogeographic groups (west, central, and eastern) with little evidence of genetic mixing. No clear phylogeographic structuring was identified in the ADMIXTURE results with more than three assumed genetic groups. Two individuals from northeast Uganda (eastern clade) showed some inferred genetic mixture with the central clade (Figure 1B). However, given its geographic distance, further analysis was required to understand the source of this introgression.

Our phylogenetic analyses largely recapitulated these three groups (and all were strongly supported as monophyletic). The first branch was the western cluster, followed by a central clade with a short branch leading to its crown and an eastern clade with a long branch. The eastern clade's first two branches were from the northern Uganda locality, which did not form a clade. Mitochondrial DNA revealed a more complex story, with two notable subclades in the eastern group. One corresponded to the Albertine Rift populations, and the other to the surprising pair of Northeast Uganda and five of six samples from the two localities in the central Congo Basin (Figure 2 and Figure S2). Additionally, we identified one western individual with mitochondrial DNA from the main central clade.

FIGURE 2.

Summary of the evolutionary history of Yellow‐whiskered Greenbul. (A) Nuclear phylogeny of Eurillas latirostris with the outgroup ( E. virens ). Nodes are marked with posterior probabilities (see legend). Tips are labelled with sample names, and northeast Uganda samples are denoted with an asterisk (*). The inset map is of sampling points with colours representing the populations these sampling localities belonged to. Country codes are as follows: DRC, Democratic Republic of the Congo; GNQ, Equatorial Guinea; GAB, Gabon; CMR, Cameroon; CAF, Central African Republic; UGA, Uganda; BDI, Burundi; SLE, Sierra Leone; GHA, Ghana. Inset tree is a summary of the mitochondrial DNA phylogeny. A full tree is in Figure S2, and cases of mitonuclear discordance are represented by concentric circles on the map, with the inner circle representing mitochondrial DNA. (B) Inferred demographic history in GADMA, with the x‐axis denoting effective population size of lineages and the y‐axis the time since divergence. Only migration edges with inferred N_m greater than 0.1 (i.e., 1 migrant every 10 generations) are shown, but the preferred model had all possible edges between modern populations.

The phylogenetic network reflected similar patterns of divergence between the genetic clusters previously identified, with the central and eastern populations broadly separated by the longest internal branch, while the western populations branch off near the base of the central group (Figure S1). The two individuals from northeast Uganda showed signatures of potential ghost ancestry, although overall recovered within the eastern group, branching off somewhat apart. This indicates that they are slightly differentiated from the remaining individuals.

3.3. Population Genetic Summary Statistics

Among the three genetic clusters, eastern and western populations were the most differentiated, with an average F_ST of 0.357 across all autosomes, as opposed to lower values when either was compared to the central one (eastern 0.218, western 0.191; Table S3). Additionally, the population from northeast Uganda was modestly (F_ST = 0.058) divergent from its nearest sampled point on Rwenzori and further divergent from other Eastern populations (0.0635; Table S3). However, this sampling locality had lower F_ST with both Western (0.300) and Central (0.153) populations, possibly reflecting its admixed nature. Genome‐wide nucleotide diversity in the main clusters was highest in the central population (0.012), followed by the western populations (0.010), then the main eastern (0.0080) population (Table S3, diagonal). Nucleotide diversity varied among eastern subpopulations, with similar diversity in the Northern Uganda populations (0.0078) to the pooled eastern populations above, with reduced diversity in the isolated Rwenzori population (0.0069). Pooling other eastern sites likely resulted in inflated diversity, as all but one eastern individual have reduced heterozygosity compared to Northern Uganda (Figure S4), and the effective diversity surface inferred higher diversity in Northern Uganda (Figure 3B).

FIGURE 3.

Visualization of population structure and genetic diversity across the landscape, with points coloured by major genetic clusters. (A) Effective migration surface, where cooler colours represent no deficit in gene flow, and warmer colours represent a deficit. Points denote sample locations and are coloured by population, separating Northeast from other Eastern populations. Points with a second, smaller point inset display mitonuclear discordance, and the inner circle denotes the mitochondrial clade (Figure 2A). (B) Effective nucleotide diversity surface, with cool colours representing lower and warm colours higher genetic diversity. Note the correspondence between transitions in genetic diversity levels and deficits in inferred migration.

Tajima's D varied notably among populations but was generally negative, suggesting population expansion. First, Tajima's D was notably more negative in core central populations than all others, including populations in central Democratic Republic of the Congo (Figure 5C), suggesting the strongest signal of recent expansion at the site of the old refugium. Second, it was reduced in the more isolated mountain populations in Central and Northern Uganda, suggesting relative stability. Interestingly, this contrasts with a study on a putatively related population of E. latirostris in Kenya, which recovered signals of an expansion in mitochondrial data (De Santis et al. 2018).

FIGURE 5.

The history of population expansion and introgression in Yellow‐whiskered Greenbul. (A) Hypothesized range of major Pleistocene forest refugia (dashed lines) and present ranges of genetic clusters (coloured blobs). Note the uncertainty in the bound between the eastern and central clusters. (B) Hypothesized history of mitochondrial capture in the Central Congo populations, based on our findings of discordance between nuclear (outer circle) and mitochondrial (inner circle) DNA. From left to right, the panels depict refugial ranges, hypothesized point of contact based on mtDNA, and finally the joint process of nuclear genetic swamping and mitochondrial capture. Figure roughly based on wavefront model presented in Currat et al. 2008. Note that this does not cover the ghost introgression inferred in the Northeast population. (C) Box plot of windowed Tajima's D per subpopulation, coloured by primary nuclear cluster (blue, west; red, central; yellow, eastern). Note that the weighted mean of Tajima's D showed an identical pattern. (D) Plot of inferred gene flow among populations using the f_branch statistic. (E) Admixture graph inferred by Treemix, with arrows representing gene flow inferred between extant or ancestral populations (as with the edge with a higher migration weight going to the Northeast population). Arrows are labelled by migration weight.

Windowed summary statistics were familiar to those in other modestly diverged taxa, with peaks in F_ST generally corresponding to shared troughs of nucleotide diversity (Figure S3). Due to the modest coverage and equal sex ratio in the dataset, we were limited in the scope we could analyse Z chromosome patterns. However, in the few comparisons we could examine, divergence was elevated and diversity was reduced (Table S3).

3.4. Demographic Histories

The best fit demographic model estimated by GADMA inferred a divergence of the western from other populations 4.27 million years in the past, followed by demographic expansion and a central‐east divergence 1.8 million years in the past (Figure 2B and Figure S5). Modern population sizes were inferred to be around three million for eastern and five million for western and central populations. Gene flow inferred to be highest from the larger central population into the western and eastern populations towards the present, with high past gene flow from the ancestral eastern/central population into the western (possibly inflating divergence times). Although independent runs converged on this pattern, there was notable variation within a given set of GADMA runs. Due to the simplicity of this model, it does not fully account for long‐term population fluctuations, such as those facilitated by Pleistocene climate fluctuations. Additionally, we caution against literal interpretation of these parameter values.

3.5. Fine‐Scale Population Structure

The effective migration surface we generated recapitulated the prior results, detecting strong barriers between our three phylogeographic groups and only one barrier within them (Figure 3A). That within‐cluster barrier separated the northeast Uganda samples from the other eastern samples but was notably weaker than the barriers between the three main groups. The effective diversity surface showed that the central populations had higher genetic diversity than the other two groups, with transitions between higher and lower diversity corresponding with areas of inferred lower migration (Figure 3B).

When focusing on the mountainous eastern populations, there was clear separation between most sampling sites. As with other sites, northeast Uganda stood out the most, but samples from the isolated Rwenzori were also quite distinct (Figure 4B). Within the core eastern samples—that is, excluding Rwenzori and northeast Uganda—we did not find any correlation between geographic or environmental distance and genetic divergence. A multiple regression model of both explained only 1.6% of the genetic differentiation (F = 0.84, R² = 0.016, p = 0.44). Neither geographic nor environmental (altitude) distance was a significant predictor when controlling for the effect of the other (p = 0.94; p = 0.96, respectively). The independent linear regressions confirmed this result (Figure 4C–D).

FIGURE 4.

Population substructure within the Eastern genetic cluster. (A) Fine scale map of sampling sites, with the base map representing elevation. The divergent central population (F_ST = 0.21) is included for reference. Inset denotes geographic context. Note that the Agoro Agu population is referred to as “Northeast” elsewhere. (B) Principal component analysis of eastern populations, with sampling locations labeling points or confidence ellipses. (C, D) Relationship between genetic distance and both geographic (C) and environmental (D) distance for a subset of eastern samples, with output from linear regression. Only samples from the main cluster of Eastern localities were included. That is, Rwenzori and Agoro Agu were excluded from this analysis.

3.6. Tests of Gene Flow

Our f_branch metric revealed genome‐wide introgression between eastern and central populations of E. latirostris , with approximately 2.7% of variation in their genomes shared due to gene flow (Figure 5D). Western populations shared small amounts of variation with central and Rwenzori populations (~0.005% and ~0.01% respectively) due to possible gene flow, although the notion that they hybridized after diverging could be spurious based on how other authors have treated f_branch metrics yielding similar values (Malinsky et al. 2021).

4. Discussion

Here we demonstrated that a forest generalist bird showed population structure consistent with the three commonly hypothesized lowland rainforest refugia in Africa (Upper Guinea, West Lower Guinea, and East Lower Guinea; Mayr and O'Hara 1986; Figure 5A) but displayed notable structure in the mountainous part of its range in the Albertine Rift. Additionally, we uncovered a complex dynamic of secondary contact, with one case of local mitochondrial capture and an inference of ghost introgression in a now‐isolated population (Figure 5). This demonstrates that taxa that currently display widespread, unspecialized ranges can track similar dynamics to different elevational specialists. That is, population structure largely aligned to lowland rainforest refugia, but more shallow divergence in the lineage in the montane region aligned with major range gaps. However, this difference may in part be driven by partial reproductive isolation of the two lineages that likely share a current contact zone in the Congo Basin.

The role of Pleistocene refugia in African forests is a topic that has seen significant attention over time, particularly due to their relevance to the evolution of Africa's unique biota and the conservation of megafauna of the region (Anthony et al. 2007; Duminil et al. 2015; Cooper et al. 2022; Ernst et al. 2025). Much of this work focuses on single habitat types, such as the lowland rainforest (Huntley et al. 2019) or highland montane forest (Kahindo et al. 2007; Mairal et al. 2017; Manthey et al. 2022). Although some work has inferred widespread montane forest (Allen et al. 2021) across the Congo Basin during the Pleistocene, the population structure demonstrated by this focal forest generalist provides support for intervening non‐forest habitat across central Africa. Additionally, two other widespread species—the Green Hylia and Yellow‐rumped tinkerbird—both show phylogeographic breaks consistent with this hypothesis (Marks 2010; Ogolowa et al. 2025).

Although our sampling scheme allowed us to find strong support for populations expanding from forest refugia, we recovered mixed support for the role of present barriers in structuring genetic diversity. One present gap that delimits both past and present forest gaps—the Dahomey Gap—separated the West from Central populations in our dataset and many other taxa in the region (Fuchs and Bowie 2015; Dongmo et al. 2019). However, we found mixed evidence for the role of riverine barriers in shaping population structure. The Congo River is spanned by both nuclear and mitochondrial genomes of the Central population in the west of the Congo Basin, but appears to have been porous enough for at least the mitochondrial DNA of the eastern population to cross (if not the whole population, as we hypothesize in Figure 5B). This river has been inferred as a mitochondrial break for some—but not all—taxa that span it (Voelker et al. 2013). However, our data provide further evidence of its porous nature, potentially suggesting a reduced role in population divergence compared to Amazonian rivers (Thom and Aleixo 2015; Musher 2025).

Our sampling spanned the focal hypothesized lowland refugia and mountain forest fragments, but we still lacked the fine scale sampling to uncover the nature of ghosts of past introgression—both the mitochondrial genome of samples in the Central Congo region and nuclear genomes of Northeast Uganda. The population in Northeast Uganda showed a surprising signal of ghost introgression from a population either sister to the Central population (as inferred by TreeMix; Figure 5E) or sister to both Central and Eastern populations (suggested by its confounding nature in tests for gene flow; Figure S6; Tricou et al. 2022; Askelson et al. 2023). This also manifested in an inferred admixed nature in population assignment analyses (Figure 1) and an elevated nucleotide diversity despite its narrow range (Figure S4 and Table S3). With our current sampling, we cannot be sure if this is the result of a formerly endemic population being swamped by the Eastern population or if it represents a hybrid population between the Albertine Rift population and the unsampled Eastern Arc population (as has been shown in island taxa; Andersen et al. 2021; Gyllenhaal et al. 2025). An affinity for Mount Morungole of Northeast Uganda and the Eastern Arc would be consistent with similarities in the avifauna on the two ranges and the relative proximity of the two ranges (Cooper et al. 2022).

The Central Congo mitonuclear discordance (strong mitochondrial but minimal nuclear introgression; Figure 5) demonstrates our incomplete understanding of a contact zone lying between these sampling points. Although EEMS infers a major deficit in gene flow in the region (Figure 3A), without contact zone sampling we cannot fully know the nature of reproductive isolation of these two populations. We believe that the asymmetry of this introgression points to some degree of nuclear reproductive isolation, and more complete sampling could examine if nuclear genes co‐introgressed with the divergent mitochondria. However, we don't know if those eastern populations in the contact zone also share mitochondrial DNA with Northeast Ugandan samples, or if this ghost mitochondrial introgression represents an even more complicated history. Additionally, it is unclear if the Congo River structures nuclear contact between the two lineages despite mitochondrial DNA spanning it. This is in contrast to Voelker et al. 2013, which found it was a mitochondrial break for four species, and work in the Amazon basin of South America where strong mitochondrial breaks are often documented (Del‐Rio et al. 2022). The best comparison for the present study is likely work by Ogolowa et al. (2025), which recovered a genetic break between samples in the Albertine rift and those to the west in another forest generalist (Yellow‐rumped Tinkerbird; Pogoniulus bilineatus ). Ogolowa et al. also found no impact of the Congo River, but rather evidence for a contact zone between the east and west tinkerbird lineages in the far eastern Democratic Republic of the Congo. However, that same region had pure eastern lineages in the present study, suggesting either a broader or further west than the in tinkerbirds (Figure 1B).

Due to our uncertainty about the dynamics of secondary contact and lack of known phenotypic differences between the Eastern and Central lineages, we opt not to make further taxonomic recommendations. However, we strongly encourage future research on this divergence, particularly because this potential reproductive isolation is within what is currently recognized as a single subspecies by some authorities (AviList Core Team 2025), although one we can find attributes the subspecies eugenius to the Eastern lineage we sampled (Lepage & Warnier 2014).

The nature of this mitonuclear discordance in tandem with the substantially higher nucleotide diversity and inferred effective population size of the Central population (Figure 2B) bears similarity to the wavefront model of asymmetric gene flow (Currat et al. 2008; Petit and Excoffier 2009). In that model, one taxon invades the range of another it is partially reproductively isolated from, then the invader experiences gene flow from the taxon that is being demographically and genetically swamped (Figure 5B). This can occur at neutral loci (and appears to have occurred for ~2.5% of the genome of Central Congo birds; Figure 5D,E), but is also hypothesized to occur for loci that disperse less. Although many birds have female‐biased dispersal, natal dispersal is a famously difficult quantity to estimate (Sutherland et al. 2000; Woltmann et al. 2012), and other taxa with male‐biased dispersal show disproportionate mitochondrial introgression (Petit and Excoffier 2009). Alternatively, this could represent the start of a mitochondrial sweep with some nuclear component (either selective or neutral), either due to positive selection on “Northeast” mitochondria in the Central population or a unidirectional mitonuclear incompatibility (i.e., Central nuclear genes are incompatible with East mitochondria, but not vice‐versa). Future work could examine the core of the contact zone to untangle alternate hypotheses of mitonuclear discordance (Rheindt and Edwards 2011; Andersen et al. 2021).

Here we have presented the complex dynamics of secondary contact following expansion from forest refugia in the Afrotropics, a region that has seen disproportionately little avian genomic research relative to its high biodiversity. Indeed, this work only tells part of the story of a cryptic contact zone between two strongly differentiated lineages in the Congo Basin. Although much work has been put into mapping refugia, this and other nuclear phylogeographic work (Ogolowa et al. 2025) in the region suggests a relatively untapped potential for fruitful study of contact zones in the core of the Congo Basin. Future work might also investigate patterns of contact in forest generalists compared to rainforest specialists, with hope of assessing the role of montane habitat connectivity in the region during periods when rainforest was absent. Finally, this provides further evidence for the importance of sampling cryptic divergences across broad distributions, in this case the second largest block of tropical rainforest in the world.

Author Contributions

J.D.M. and B.D.M. conceptualized the study. J.E.C.‐P. and S.S.B. performed lab work. B.D.M., B.M., C.K. and F.B. performed field work. E.F.G., S.S.B., J.E.C.‐P., M.C.T., A.A.R. and J.D.M. conducted bioinformatic analyses. E.F.G. and B.D.M. wrote the first draft of the manuscript. All authors contributed intellectually and to the writing of the final manuscript.

Funding

This work was supported by National Science Foudnation (NSF) grants 1953688 to JDM and 1953796 to BDM. EFG was supported by the NSF grant DBI‐2410565.

Disclosure

Benefits Generated: This work was made possible by field work from several countries. The collaborators most critical to the intellectual development of this manuscript and logistical support that allowed the novel sampling for this work from the Democratic Republic of the Congo are included as co‐authors. Support from their institutions for assistance in obtaining research permits and export documentation is also recognized in the acknowledgements. More broadly, benefits from this research arise from the availability of museum specimens collected for this work and genetic data, the latter of which is publicly accessible as described in Data Availability.

Ethics Statement

All samples were collected during general collecting trips conducted by natural history museums, work that complied with local permits and ethical guidelines outlined by Institutional Animal Care and Use Committees for the institutions.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Distance‐based Neighbour‐Net phylogenetic network of the Yellow‐whiskered Greenbul, Eurillas latirostris generated with SplitsTree. Colours match populations and tips are labelled with the country code followed by the sample museum number.

Figure S2: Phylogenetic tree of mitochondrial gene ND2. Bars across tips denote populations, with some noting discordant individuals or more precise sampling localities. Inset map shows sampling localities of different populations, with mitonuclear discordance noted as two concentric circles, with mtDNA as the inner and nuclear background as the outer circle.

Figure S3: Summary statistics for population structure (Reich's F_ST, left) and nucleotide diversity (π, right) in 100 kb windows. The Y axis represents the respective statistics, and X is chromosomal position. Chromosomes are delimited by the background colour, starting with 1 on the left and 33 on the right. Note that the Z chromosome was excluded due to the relatively high proportion of females in the dataset, but F_ST was elevated and diversity reduced in the one comparison with sufficient samples.

Figure S4: Violin plot of mean genome‐wide heterozygosity (i.e., the proportion of sites that are heterozygous in a single individual) per population. Colours match populations and are labelled on the X axis. Points correspond to individuals' values. Letters are compact letter displays of Tukey's Honestly Significant Difference test (i.e., different letter denotes a significant difference, same letter denotes no significant difference). Note that the Northeast Uganda population is split from the rest of the eastern sampling localities.

Figure S5: GADMA model fit output. (A) Observed 2D site frequency spectra (SFS), (B) modelled 2D SFS, (C) residuals as a 2D SFS, and (D) distribution of residuals.

Figure S6: F_branch statistics among E. latirostris populations with northeastern Ugandan samples included as a distinct population. As in Figure 5, we used DSuite to calculate f_branch statistics per chromosome and then averaged these values across the entire genome. Darker colours indicate higher proportions of introgression among extant E. latirostris populations and internal phylogenetic branches.

Figure S7: Treemix admixture graphs for (A) zero, (B) one, and (C) two migration edges (as in Figure 5D). Arrows represent gene flow inferred between extant or ancestral populations.

Table S1: Sampling and locality information for all individuals. Columns from left to right are sample ID (tissue # for KU specimens), species, nuclear DNA cluster, mtDNA cluster, sampling country, latitude, longitude, sex, mean read depth, number of sites genotypes, and number of clean reads. FMNH = Field Museum of Natural History. KU = University of Kansas Biodiversity Institute. Full sampling table with additional information (sample name, original sex ID, elevation, state/region) can be found in github and dryad uploads.

Table S2: Select datasets used in this study and their characteristics, focusing on dataset that highlight the more unique combinations of missingness, outgroup inclusion, and thinning parameters.

Table S3: Table of summary statistics for focal populations. The numbers below the diagonal correspond to autosomal F_ST, above the diagonal to Z chromosome F_ST for populations with enough males, and the diagonal to nucleotide diversity (pi). Bolder lines outline major clusters (West, Central, and East).

Data Availability Statement

New data generated here will be uploaded to NCBI SRA PRJNA1425520. Code is available at https://github.com/ethangyllenhaal/YellowWhiskeredGreenbul/. This code and select input and output files can be found at https://doi.org/10.5061/dryad.b8gtht7sx.